LeXtract.Schema (lextract v0.1.2)

Generate and manage NimbleOptions schemas from extraction examples.

Provides schema generation for structured output with ReqLLM's generate_object/4.

Example

iex> examples = [
...>   %LeXtract.ExampleData{
...>     input: "Patient takes aspirin",
...>     output: %{
...>       "extractions" => [
...>         %{"class" => "Medication", "medication_attributes" => %{"name" => "aspirin"}}
...>       ]
...>     }
...>   }
...> ]
iex> schema = LeXtract.Schema.from_examples(examples)
iex> is_list(schema)
true
iex> Keyword.has_key?(schema, :extractions)
true

Summary

Functions

from_examples(examples, opts \\ [])

Generates NimbleOptions schema from example extractions.

merge(generated_schema, user_schema)

Merges user-provided schema with generated schema.

validate(data, schema)

Validates extraction data against a schema.

Functions

from_examples(examples, opts \\ [])

@spec from_examples(
  [LeXtract.ExampleData.t()],
  keyword()
) :: keyword()

Generates NimbleOptions schema from example extractions.

Analyzes the provided examples to automatically infer a schema suitable for use with ReqLLM's generate_object/4. The schema will include all extraction classes and their attributes found in the examples.

Parameters

examples - List of LeXtract.ExampleData structs
opts - Options (see below)

Options

:attribute_suffix - Suffix for attribute fields (default: "_attributes")
:required - List of required field names (default: [])
:manual_schema - User-provided schema to merge with generated (default: nil)

Examples

iex> examples = [
...>   %LeXtract.ExampleData{
...>     input: "Dr. Smith prescribed aspirin to John Doe",
...>     output: %{
...>       "extractions" => [
...>         %{
...>           "class" => "Person",
...>           "person_attributes" => %{"name" => "Dr. Smith", "role" => "doctor"}
...>         },
...>         %{
...>           "class" => "Medication",
...>           "medication_attributes" => %{"name" => "aspirin"}
...>         }
...>       ]
...>     }
...>   }
...> ]
iex> schema = LeXtract.Schema.from_examples(examples)
iex> Keyword.has_key?(schema, :extractions)
true

merge(generated_schema, user_schema)

@spec merge(keyword(), keyword()) :: keyword()

Merges user-provided schema with generated schema.

User schema takes precedence for conflicts. This allows users to override specific parts of the automatically generated schema while keeping the rest.

Parameters

generated_schema - Schema generated from examples
user_schema - User-provided schema overrides

Returns

Merged schema as keyword list.

Examples

iex> generated = [
...>   extractions: [type: {:list, [type: :map]}, default: []]
...> ]
iex> user = [
...>   extractions: [type: {:list, [type: :map]}, default: [], required: true]
...> ]
iex> merged = LeXtract.Schema.merge(generated, user)
iex> merged[:extractions][:required]
true

iex> generated = [
...>   name: [type: :string],
...>   age: [type: :integer]
...> ]
iex> user = [
...>   age: [type: :integer, required: true]
...> ]
iex> merged = LeXtract.Schema.merge(generated, user)
iex> Keyword.has_key?(merged, :name)
true
iex> merged[:age][:required]
true

validate(data, schema)

@spec validate(
  term(),
  keyword()
) :: {:ok, term()} | {:error, term()}

Validates extraction data against a schema.

Uses NimbleOptions.validate/2 internally to validate the provided data against the schema.

Parameters

data - Data to validate (typically a keyword list or map)
schema - NimbleOptions schema (keyword list)

Returns

:ok if valid, {:error, reason} otherwise.

Examples

iex> schema = [
...>   extractions: [
...>     type: {:list, :map},
...>     default: []
...>   ]
...> ]
iex> data = [extractions: []]
iex> LeXtract.Schema.validate(data, schema)
{:ok, [extractions: []]}

iex> schema = [
...>   extractions: [
...>     type: {:list, :map},
...>     required: true
...>   ]
...> ]
iex> data = []
iex> {:error, %NimbleOptions.ValidationError{}} = LeXtract.Schema.validate(data, schema)